ARTICLE IN PRESS

Int. J. Human-Computer Studies 66 (2008) 3650 www.elsevier.com/locate/ijhcs

A social network-based system for supporting interactive collaboration in knowledge sharing over peer-to-peer network
Stephen J.H. Yanga,, Irene Y.L. Chenb
a

Department of Computer Science and Information Engineering, National Central University, No. 300, Jhongda Road, Jhongli City, Taoyuan County 320, Taiwan b Department of Information Management, Ching Yun University, No. 229, Chien-Hsin Road, JhongLi 320, Taiwan Received 24 October 2006; received in revised form 15 August 2007; accepted 29 August 2007 Communicated by C. Sierra Available online 4 September 2007

Abstract Knowledge sharing enables people in virtual communities to access relevant knowledge (explicit or tacit) from broader scope of resources. The performance in such environments is fundamentally based on how effectively the explicit and tacit knowledge can be shared across people, and how efciently the created knowledge can be organized and disseminated to enrich digital content. This study will address how to apply social network-based system to support interactive collaboration in knowledge sharing over peer-to-peer networks. Results of this study demonstrate that applying such social network-based collaboration support to knowledge sharing helps people nd relevant content and knowledgeable collaborators who are willing to share their knowledge. r 2007 Elsevier Ltd. All rights reserved.
Keywords: Social network; Interactive collaboration; Knowledge sharing; Peer-to-peer; Instant messenger

1. Introduction The explosion in Web-based technology has led to increasing volume and complexity of knowledge, which stimulates the proliferation of the virtual communities (e.g., peer-to-peer (P2P) network and instant messenger (IM)). The Internet provides an environment for people to contribute their knowledge and acquire others. One of the Internets intended purposes is to encourage knowledge sharing so that valuable knowledge embedded in the network can be effectively explored. Most Internet users expect that they can acquire and share valuable knowledge to fulll their needs. However, knowledge sharing in some virtual communities has not lived up to the expectation. Two barriers preventing efcient and effective knowledge sharing are (Chen and Yang, 2006): (1) The difculty in nding relevant knowledge. (2) The difculty in nding relevant collaborators to interact with.
Corresponding author. Tel.: +886 3 422 7151; fax: +886 3 422 2681.

E-mail addresses: jhyang@csie.ncu.edu.tw (S.J.H. Yang), irene@cyu.edu.tw (I.Y.L. Chen). 1071-5819/$ - see front matter r 2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.ijhcs.2007.08.005

Knowledge sharing requires collaboration between the consumers and contributors of knowledge; namely the collaborators. This task cannot be accomplished simply by storing knowledge in the repository. It also requires a mechanism, which helps people nd the collaborators with relevant knowledge. Collaboration over the Internet communities has characterized itself by heavily relying on interaction among the collaborators (Bistrom, 2005; Eikemeier and Lechner, 2003). Collaborators can be any virtual users who interact to achieve the goals of resources discovery, access, knowledge sharing, group communication and discussion. The collaboration for knowledge sharing should be enacted without spatial and temporal limitations. In addition, it should take place over medium such as the Internet and therefore beyond the geographical boundaries. This research studies how to enhance knowledge sharing through a social network-based collaboration support. Social networks are built upon an idea that there exists a determinable networking structure of how people know each other. In such networks, people are connected through common social relationships either directly or

ARTICLE IN PRESS
S.J.H. Yang, I.Y.L. Chen / Int. J. Human-Computer Studies 66 (2008) 3650 37

indirectly (Churchill and Halverson, 2005). Social networks identify two important relationship tiesknowledge relationship tie and social relationship tie. The two relationship ties are metrics used to measure the degree of peoples knowledge matching with a query, and the degree of peoples social relationship with other people. Our social network-based collaboration support is implemented with P2P and IM technology to enhance content discovery and interactions among collaborators in knowledge sharing. P2P makes each peer act as both client and server (Aberer et al., 2002; Gnutella). It provides a metaphor commonly addressed in virtual communities that an individual could be a consumer as well as a producer of knowledge. This makes P2P particularly suitable for facilitating efcient knowledge sharing in Internet communities. P2P collaboration is supported by IM (Herbsleb et al., 2002; Isaacs et al., 2002a), which provides real-time interaction and communication. IM reveals not only how many people are online at a given time, but also who is online (Alexander, 2005; Grinter and Palen, 2002; Isaacs et al., 2002b; Nardi et al., 2000). P2P, IM, and social networks share many concepts in common. For example, they are both distributed networking structures; a peer in a P2P network is like a node in a social network; a link in a P2P network is like a relationship tie in a social network. The major contribution of this paper is that it provides a mechanism, which adopts our social network-based system in the domain of a campus-wide P2P network to overcome the two aforementioned barriers. The studies and experiments presented in this paper demonstrate that applying our social network-based system can help people nd relevant and quality resources over a campus-wide P2P network as well as relevant people who are willing and capable of sharing their knowledge. For nding relevant and quality resources, we utilized a rating mechanism to calculate a resources reputation (Mitchell, 1997). For nding relevant people, we developed social net and knowledge net to calculate the degree of relationship ties between people in order to nd the most related people who are willing and capable of sharing their knowledge (resources). In detail, we take a three-step approach. (1) Firstly, we construct a P2P network of content repositories with 156 peers in a university. (2) Secondly, upon receiving a request of collaboration from a user, we dynamically generate a network depending on the knowledge retrieved from the P2P network. (3) Finally, we create an instance messenger (IM)-based virtual community to enable real-time collaborations between peers identied by the constructed network. The remainder of the paper is organized as follows. In Section 2, we review related work. In Section 3, we introduce how to nd relevant knowledge via our constructed P2P network. In Section 4, we present how to nd relevant and knowledgeable collaborators via our

social network-based system. In Section 5, we discuss IMenabled group collaboration and analyze experimental results. We conclude this study with future research in Section 6. 2. Related work 2.1. Knowledge sharing and virtual communities The Internet gives rise to virtual communities that aim at facilitating collaboration by providing an environment for mutual sharing and interaction. A collaborative process in such an environment involves intensive online knowledge discovery and knowledge sharing between collaborators, such as knowledge consumers and knowledge contributors. The emergent virtual communities over the past decade have stimulated research interests by academia and practitioners. For example, Zhang and Tanniru (2005) built an agent-based model for virtual learning communities (VLCs). The results of a series of comparison experiments indicated that every member tends to be active in the initial stage of the development of a VC; when the community reaches a reasonable size, each participant is important and function together to form a healthy and stable population. Bruckman (2002) found that the potential of Internet technology could come from the peers and elders. Jin (2002) provided a conceptual framework for the development of a prototype system of the virtual community based interactive environment. Wachter et al. (2000) pointed out that an enhanced environment is possible only if one goes beyond mere on-line course delivery and creates a community of people and other related resource groups. Wasko and Faraj (2005) found that knowledge sharing has been a motivation for participation in virtual communities. Although prior studies have provided evidence demonstrating the importance of knowledge sharing in enhancing the performance, few have attempted to provide mechanisms to support the knowledge sharing in virtual communities. 2.2. Collaboration support and social networks One of the important supports in collaboration systems is the awareness of collaborators and their actions (Carroll et al., 2003; Diamadis and Polyzos, 2004). For example, who are the collaborators? What do they know? And what are they doing now? Fischer et al. (2005) claimed that peoples creativity is the result of interaction and collaboration with other individuals in a society. In other words, creativity is a social activity that embodies in collective knowledge and essential collaborators. This indicates that nding knowledge and nding collaborators are two important supports for knowledge sharing in collaboration systems. Besides the supporting collaboration tools, one factor critical to the success of knowledge sharing in virtual communities is volunteered knowledge contribution, which

ARTICLE IN PRESS
38 S.J.H. Yang, I.Y.L. Chen / Int. J. Human-Computer Studies 66 (2008) 3650

is inuenced by the social relationships in a network. Social relationships are composed of nodes and ties. Nodes are the individual actors within the network, and ties are the links between the actors. There can be many kinds of ties between the nodes. In its most simple form, a social network is a map of all of the relevant ties between the nodes being studied. Researchers have recognized that a broader sense of social network is a self-organized structure of people, information, and communities (Kautz et al., 1997; Raghavan, 2002). A social network can be modeled by a net structure consisting of nodes and edges. Nodes represent individuals or organizations. The edges connecting nodes are called ties, which represent the relationships between the individuals and organizations. The strength of a tie indicates how strong the relationship is. Many kinds of ties may exist between nodes. In this paper, we will address the knowledge relationship tie and the social relationship tie. Social networks can be represented as matrices; therefore, the properties of the social networks can be analyzed by graph theory. Our social network-based P2P network is designed based on Hill and Dunbars theory (Hill and Dunbar, 2002). They suggest that the maximum size of a social network averages around 150 individuals. Thus, we conne the number of nodes in our social network-based P2P network to around 150. Another good reason to justify the number of nodes in our P2P network is that small communities are more prone to collaborate than large ones. The other two reasons to conne the number of nodes in our P2P network to around 150 on-campus nodes are concerns of copyright and message ooding. Firstly, to reduce the possibility of misusing copyrighted digital content that do not belong to the university (or is not subscribed to the university), we conne the users of our system to the nodes with on-campus IP address. The idea is similar to book publishers policy that only the end users who search for information via on-campus IPs are allowed to access subscribed digital content. The other reason to set an upper bound of around 150 nodes is to avoid the socalled ooding phenomenon (Ravishankar, 2002). Flooding can be dened by the equation
TTL X i1

to 150 to avoid the ooding phenomenon in current experiment. In our research, the scope of a social network is limited to 156 peers in a university. 2.3. Peer-to-peer and instant messenger Two of the very good examples of virtual communities are P2P and IM. P2P is a connected network of content repositories, which is referred to as one of the resources of accessible explicit knowledge. IM enables interactive group collaboration through which people can interactively communicate, discuss, and exchange ideas to derive tacit knowledge. The discussion and interaction details recorded in IM are considered as one of the resources of tacit knowledge. Besides the well-known Gnutella protocol for le search and le sharing, many researchers have applied P2P to collaboration. An academic P2P framework- Edutella, was developed for learning communities (Brase and Painter, 2004; Nejdl et al., 2002). This framework uses the exchange of Resource Description Framework (RDF) metadata to enhance the description and discovery of learning resource. In contrast to conventional P2P keyword matching, Edutella proposed a semantic matching approach. Each peer on the Edutella P2P network maintains a data repository, which serves as a service registry containing metadata to facilitate the description and discovery of learning resource. Bistrom (2005) suggested using the P2P network in collaborative learning environment. In his study, Bistrom addressed certain essential requirements for successful collaborative learning, and how P2P networks should be developed with additional communication functionalities to better support collaborative learning. Eikemeier and Lechner (2003) proposed a P2P collaboration tooliKnow, for domain specic ad-hoc collaboration. iKnow enhanced a typical P2P platform with IM communication support in order to foster communication among P2P communities. iKnow extended P2P search functionality to locate experts who are currently on the P2P network, and then establish an IM network among the experts. Groove (2006) is a commercial P2P collaboration support tool with which users can join various groups based on interest and expertise. The communication among users can be encrypted, but this results in computation overhead. Unlike most academic P2P platforms, users of Groove are not anonymous; they need to register and identify themselves, which also results in administration overhead. P2P and social networks share many concepts in common. For example, they are both distributed networking structures; a peer in a P2P network can be viewed as an analog of a node in a social network; a link in a P2P can be viewed as an analog of a relationship tie in a social network. In contrast to most P2P searches that emphasize on search queries and protocols, our social network-based P2P search aims at reducing search time and decreasing message trafc by minimizing the number of messages circulating in the network.

ni

nnt 1 . n1

For an n-nary node in a P2P network, its centrality is n. TTL is the abbreviation of time to live, which is the period of time during which a request is valid. A request will become invalid when the TTL is expired. If the TTL is too short, a request may turn invalid before it can be satised. If the TTL is too long, a ooding phenomenon is prone to occur when the request cannot be satised in a short time. The values of n and TTL depend on the total number of nodes in a P2P network, thus the more nodes, the larger n and longer TTL, which will lead to potential ooding phenomenon. As a result, we conne the number of nodes

ARTICLE IN PRESS
S.J.H. Yang, I.Y.L. Chen / Int. J. Human-Computer Studies 66 (2008) 3650 39

The goal of this research is to overcome the aforementioned two barriers that hinder efcient and effective knowledge sharing. In the following two sections, we will address how to overcome the two barriers by classifying knowledge and dening its quality control mechanism, and classifying collaborators and dening their relationship ties. Based on the classications and denitions, we present the implementation of our P2P and IM collaboration support environment and demonstrate how to utilize such environment to nd relevant knowledge and right collaborators. 3. Finding relevant knowledge via P2P network Bieber et al. (2002) pointed out that a virtual communitys knowledge has both explicit and tacit components. Explicit knowledge is mostly in the forms of documents and tangible artifacts that can be expressed in words, languages, diagrams, and formulas (Polanyi, 1997). In contrast, tacit knowledge is mostly in the forms of concepts and intangible personal experiences. Explicit knowledge can be codied, represented and shared asynchronously via reading and studying documents, while tacit knowledge cannot be explicitly described and mostly can only be perceived in human minds and shared synchronously via discussion and mentoring (Gall, 1987). Examples of explicit knowledge include books, research papers, and video recordings. Examples of tacit knowledge include know how, cognitive skills, and beliefs. In this paper, we refer to content found by P2P as one of the resources of explicit knowledge, while the discussion and interaction

details recorded in IM is one of the resources of tacit knowledge. 3.1. Architecture of P2P network To facilitate knowledge discovery, we have constructed a P2P network. As shown in Fig. 1, each peer in the P2P network consists of two modules: Resource Module and Search Module. The Resource Module is designed to formally describe resources contained in a peer. The Search Module is responsible for generating users search queries and processing their search requests. The Resource Module contains several managers to organize and manage the resources kept in the peer. The Resource Manager is the coordinator, which handles all kinds of resources from various managers. These resources can be content, services, or other applications provided by the peer. The managers include the Content Manager that handles the content repository, the Ontology Manager that provides semantic metadata of contents, and the Annotation Manager that processes annotation imposed to the content. The Search Module contains a Query Process Engine and a Resource Discovery Manager. The Query Process Engine is an interface designed to generate search request. If a user cannot specify a search request clearly, the Query Process Engine automatically generates one for him/her based on his/her surrounding context. The Resource Discovery Manager is designed to process search requests received from peers by providing a concept map to guide the search process. The concept map is derived from the

Fig. 1. System architecture of peer-to-peer network.

ARTICLE IN PRESS
40 S.J.H. Yang, I.Y.L. Chen / Int. J. Human-Computer Studies 66 (2008) 3650

keyword and keyword thesaurus analyzed based on users requests; the concept map is then extended or redrawn to match users search requests. 3.2. P2P-based content discovery For content discovery, our P2P network provides the functions of basic keyword search, keyword thesaurus and concept map-based search. Based on the content classications and their quality control, the keyword thesaurus is used to extend search scope by nding more relevant keywords. In contrast, concept map based search is used to derive a more precise search scope by nding the most relevant keyword. As shown in Fig. 2, the basic keyword search is enhanced by utilizing the keyword thesaurus. Our P2P network matches not only a single keyword but also a set of related keywords previously classied and saved in our content repository. The search results are shown in the main window along with the le name, type, size, state, and rating of the resource. For example, a keyword search of New York Vacation will derive a keyword thesaurus such as New York City Life, New York Travel, and even New York Yankee. For semantic search, we utilize the concept map approach to construct the relationship of a keyword concept and its related concept (Chau and Yeh, 2004). For example, if a user inputs the concept Web services,

the system will produce a concept map with three nodes and two edges. One edge connects Web services and Semantic Web, and the other connects Web service and DAML-S. If the user continues to press the node Semantic Web, the concept map will grow further to the one shown in Fig. 3. If the user then double clicks the node XML, the system will proceed to search and come out with the results. On the upper left side of Fig. 3, there is the description of the concept. The lower left side of Fig. 3 shows the types of resources and their abstracts related to the concept, while the lower right side shows the details of the resource selected from the lower left side window. 3.3. Enhancing knowledge discovery with quality control The motivation for participating in virtual communities is knowledge sharing (Wasko and Faraj, 2005). Without high-quality knowledge as its content, a virtual community cannot achieve its intended purpose of encouraging knowledge sharing. Content resource management can be used as a mean to achieve this goal. To facilitate content resource management, we classify resources based on their knowledge domains and their quality. A peer can have a variety of resources and a single resource can be replicated in more than one peer. We utilize ACM Computing Classication System (1998) as our classication base of knowledge domain. In order to organize and provide better resource management, each

Fig. 2. P2P network with keyword thesaurus search.

ARTICLE IN PRESS
S.J.H. Yang, I.Y.L. Chen / Int. J. Human-Computer Studies 66 (2008) 3650 41

Fig. 3. P2P network with concept map.

peer in our P2P network needs to classify the content and evaluate its quality of content based on their reputation, accessed frequency (per day), and the matching degree to which the content classication conforms to knowledge domain. In our previous study, we have utilized a rating mechanism to evaluate Web services reputation (Yang et al., 2006). In this study, we utilize the same rating mechanism to calculate a resources reputation based on a 95% condence interval in terms of probability (Mitchell, 1997), we will explain this in more details when we address a peers reputation in later section. The quality of resource i in knowledge domain j is given as QoRi;j REPi;j TOAi MDi;j , where QoR is the quality of a content resource, REP the reputation represents the rating of the resource. The higher the reputation, the better the rating of the resource, TOA the total number of times a resource is accessed per day. TOA represents the degree of popularity. The higher the TOA, the more popular the resource, MD the matching degree to which a content classication conforms to knowledge domain. The higher the MD, the better the matching. The quality of a peer with respect to a certain knowledge domain, j, is the summation of quality of resource i over

the number of content resources, as given below PNoR QoRi;j , QoPj i1 NoR where QoP the quality of a peer, NoR the number of content resources, which represents the volume of content in a peer. The quality of a peer with respect to all knowledge domain contained in this peer is the average of QoPj, which is given as follows: PNoD j1 QoPj QoP QoP NoD PNoD PNoR j1 i1 REPi;j TOAi MDi;j =NoR , NoD where NoD is the number of knowledge domains, which represents the scope of this peers knowledge. Based on the content classications and the quality control, our P2P environment can be enabled to nd more relevant and quality contents. In order to encourage peers to continually contribute quality content, each day our system will rank the peers according to their scores of reputation. The top 10% peers will be called honorary peers. The underlying concept of honorary peers is that peers making more contribution deserve higher priority to

ARTICLE IN PRESS
42 S.J.H. Yang, I.Y.L. Chen / Int. J. Human-Computer Studies 66 (2008) 3650

download content from the network. In other words, we value the honorary peers by allocating them higher download priority. 4. Finding relevant collaborators via social network The key idea of our three-layer social network is illustrated in Fig. 4. For a given query requesting for collaborators with certain knowledge, a social network containing relevant collaborators with the requested knowledge will be dynamically constructed within the scope of a P2P network. As shown in Fig. 4, the rst layer is the P2P knowledge net (K-net), which is established to incorporate peers who own the requested knowledge into a pool of active peers. In this research, we conne the scope of the pool to a smallscale P2P network with about 50150 peers within a university. A peer in the pool can be either a knowledge repository or a knowledgeable individual. A weighted edge between two peers is a knowledge relationship tie, which is used to measure the degree to which a peers knowledge matches the others query. For example, if a peers query in such a P2P K-net (e.g., peer Steve) is requesting for peers with Software Engineering (SE) knowledge, a P2P K-net will be dynamically generated based on the query. As shown in Fig. 4, there are four edges dynamically generated on P2P K-net based on Steves query. Among them, the edge weighted (0.8) connecting Chris and Steve indicates that Chriss knowledge matches best with Steves query. The edge weighted (0) connecting Mary and Steve indicates that Marys knowledge does not match Steves query. P2P social net (S-net) is the second layer. A weighted edge connecting two peers is a social relationship tie, which is used to measure the degree of social familiarity between the two peers. Peers on K-net without the requested

knowledge will be removed from S-net (e.g., Mary). Using the example shown in Fig. 4, Steve is more familiar with Albert than Chris because the social relationship tie between Steve and Albert is (0.9), which is greater than the social relationship tie between Steve and Chris (0.8). Based upon the generated S-net, an IM enabled group collaboration, shown on the third layer in Fig. 4, is invoked to help Steve communicate with the peers found in Layer-2 (Chris and Albert). Peers that appear on S-net and have negative relationships with the requester will be removed from the group collaboration (e.g., Bob). This example shows that the essential challenge of constructing this three-layer social network is how to calculate knowledge relationship tie and social relationship tie. 4.1. Calculation of knowledge relationship tie In our research, we consider a peers knowledge domain, prociency, and reputation of contribution as key indicators determining its capability to participate in collaborations. Therefore, as shown in K-net, we calculate a peers knowledge relationship tie based on the three indicators. As addressed by Barak and Rafaeli (2004), one of the most common hierarchies for ranking knowledge is the Bloom taxonomy (Bloom et al., 1956). In this study, we use the revised Bloom taxonomy in the form of matrix (Anderson et al., 2001) to classify a peers domain knowledge and its prociency in such a domain. As shown in Fig. 5(a) and (b), Bloom taxonomy is a matrix consisting of two dimensions: Knowledge dimension and Cognitive Process dimension. Knowledge dimension indicates the types of knowledge; Cognitive Process dimension indicates cognitive processing of knowledge. Each cell in a Bloom matrix is associated with a value ranging from 0 to 1,

Fig. 4. A three-layer social network.

ARTICLE IN PRESS
S.J.H. Yang, I.Y.L. Chen / Int. J. Human-Computer Studies 66 (2008) 3650 43

Knowledge dimension Factual knowledge Conceptual knowledge Procedural knowledge Metacognitive knowledge

Cognitive Process Dimensions Understand Remember 0.8 0.9 0.3 0.3 0.6 0.5 0 0

Apply 0.4 0.3 0.3 0

Analyze 0.4 0.1 0.2 0

Evaluate 0 0 0 0

Create 0 0 0 0

Knowledge dimension Factual knowledge Conceptual knowledge Procedural knowledge Metacognitive knowledge

Remember 0.2 0.8 0.2 0

Cognitive Process Dimensions Understand Analyze Evaluate Apply 0.7 0.7 0.7 0.8 0.7 0.8 0.7 1.0 0.2 0.3 0.2 0 0 0 0 0

Create 0.8 0.7 0 0

Fig. 5. (a) Example of Alberts Bloom taxonomy matrix. (b) Example of Chriss Bloom taxonomy matrix.

representing the level of prociency. The values are manually given by Bloom experts after conducting evaluation tests for classifying individuals prociency toward Knowledge dimension and Cognitive Process dimension based on the Bloom taxonomy (Anderson et al., 2001; Bloom et al., 1956). For example, let Figs. 5(a) and (b) indicate peer Albert and peer Chriss knowledge prociency regarding the knowledge domain of Software Engineering, respectively. In Fig. 5(a), the cell [Factual knowledge, Remember] has a value (0.9), which indicates that Albert is good at memorizing factual knowledge about Software Engineering. In Fig. 5(b), the cell [Conceptual knowledge, Apply] has a value (1.0), which indicates that Chris is excellent at applying conceptual knowledge of Software Engineering. 4.1.1. Generation of Bloom matrix Generally speaking, the domain experts and the Bloom experts are different individuals. In this paper, the domain expert is the course instructor who happened to be the rst author of this paper, and the Bloom expert is the second author of this paper. The evaluation questions (in the format of multiple choice) used for the generation of a Bloom matrix are given by the domain experts while they prepare examination papers for quiz, midterm or nal examinations. The collection of evaluation questions can be further organized into a test bank. At this question generation stage, the evaluation questions have not been classied into categories based on Bloom matrix (i.e., cells in a two-dimensional Bloom matrix, e.g., [Factual knowledge, Remember]). In this approach, all of the examination papers (containing the Bloom evaluation questions) can be generated from the test bank, thus there is no extra efforts for the instructor (domain expert). Students take the same copies of examination papers at the same time for both Bloom evaluation test and other regular examinations (the quiz, midterm and nal examinations), thus there is no additional work for students. The only overhead comes from the efforts of Bloom experts, who need to manually classify each evaluation

question into an appropriate category. After this stage, each evaluation question is associated with a category. For example, a Bloom matrix (Fig. 5(a)) associated with the domain knowledge of software engineering. There are two examples of evaluation questions Q1 and Q2 as shown in the following. Both Q1 and Q2 are in the domain of software engineering. Q1 will be classied by a Bloom expert into the category of [Factual knowledge, Remember] because this question is used to evaluate whether students can memorize the software process model. Q2 will be classied into the category of [Conceptual knowledge, Apply] because this question is used to evaluate whether students can apply the concept of waterfall approach to certain real-world applications. The number of questions in each category may be different depending on the number of questions given by the domain expert. Thus, the nal scores of the evaluation test need to be normalized and the results will range from 0 to 1. The calculation and the normalization of the values in a Bloom matrix are performed automatically. In our approach, each student not only gets a test score, but also gets a Bloom matrix classication when he/she nishes each test. The values in some cells of the Bloom matrix will be zero if a student scores 0 point in related categories or if there is no evaluation questions in those categories.

Q1. What follows is NOT a software process model? (1) the waterfall approach (2) evolutionary development (3) verication and validation (4) formal transformation. Q2. Given the following systems, which one is the most appropriate to be developed by applying the waterfall approach? (1) an anti-lock braking control system (2) a virtual reality system (3) an accounting system (4) an interactive system.

ARTICLE IN PRESS
44 S.J.H. Yang, I.Y.L. Chen / Int. J. Human-Computer Studies 66 (2008) 3650

4.1.2. Knowledge relationship tie Huhns and Buell (2002) pointed out that people are more likely to trust something proved. Similarly, people are more likely to collaborate with someone with good reputation. Many approaches have been proposed to calculate degree of trust (reputation) based on experiences, such as rating mechanism and referral networks. In our previous research, we developed a rating mechanism to evaluate Web services reputations. The mechanism applied sampling of binomial probability to calculate a services reputation based on a 95% condence interval in terms of probability (Mitchell, 1997). The binomial distribution approximates the normal distribution when the number of samples is large enough. For the normal distribution, the condence interval is 95% probabilities falling in the range of mean71.96 SD (Standard Deviation) in compliance with the experience rule. We applied the same rating mechanism in this research to evaluate peers reputation. Due to length limitation, readers please refer to (Yang et al., 2006) for detailed discussions about the calculations of trust and reputation. Consider that peer is query is requesting for peer j whose knowledge prociency conforms to a requested knowledge domain k. Peer is query can be calculated by KQk i; j K proficiency jK conformance iT , k k KQserialized i; j k
4 6 X X m1 n1

on the aforementioned equations and the example shown in Figs. 4 and 5, we found two peers, Albert and Chris, whose knowledge relationship ties, K tie Steve; Albert and SE K tie Steve; Chris are greater than zero, which means both SE Albert and Chris conform to Steves request in terms of knowledge relationships. Moreover, since K tie Steve; SE Chris is greater than K tie Steve; Albert, Chris is more SE knowledgeable than Albert in terms of meeting Steves request of Applying conceptual knowledge of Software Engineering. The detail calculation of knowledge relationship tie is illustrated as follows: Let Steve0 s request be K conformance SteveT SE 3T 2 0 0 0 0 0 0 7 6 60 0 1 0 0 07 7 6 6 7 . 60 0 0 0 0 07 5 4 0 0 0 0 0 0 Let Alberts knowledge prociency, K proficiency Albert be SE the matrix shown in Fig. 5(a), then KQSE Steve; Albert K proficiency AlbertK conformance SteveT SE SE 32 2 0 0 0 0 0:9 0:8 0:4 0:4 0 0 76 6 6 0:3 0:3 0:2 0:1 0 0 76 0 0 1 0 76 6 6 76 6 0:6 0:5 0:3 0:2 0 0 76 0 0 0 0 54 4 2 0 0 0 0 0 0 0 0 0:2 0 0 0 0 0 0 3 0 0 0 7 0 0 07 7 7. 0 0 07 5 0 0 0 0 0 0 0 3T

! KQk m; n ,

0 0 0 0

K tie i; j KQserialized i; jK reputation j, k k k where KQk i; j is a Bloom taxonomy matrix representing a query by peer i requesting for peer j whose knowledge prociency conforms to a requested knowledge domain k. K proficiency j is a Bloom taxonomy matrix representing k peer js knowledge prociency with respect to a requested knowledge domain k. K conformance i is a Bloom taxonomy matrix representing k a conformance requirement requested by peer i to peers whose knowledge prociency conforms to a requested knowledge domain k. KQserialized i; j is the serialization of KQk i; j, which is a k 6 by 4 Bloom taxonomy matrix. K tie i; j is a real number between 0 and 1 representing k the knowledge relationship tie between peer i and peer j with respect to knowledge domain k. K reputation j is a real number between 0 and 1 representk ing peer js reputation regarding contribution to the requested knowledge domain k. The value of K tie i; j indicates the degree of knowledge k match up between peer js knowledge and peer is query; the higher the value, the stronger the tie. For example, consider a peer, Steve, who requests for peers who have knowledge Software Engineering and the prociency of applying conceptual knowledge of Software Engineering. Steves request can be denoted as K conformance Steve. Based SE

7 07 7 7 07 5 0

6 60 6 6 60 4 0

After serialization KQserialized Steve; Albert k

4 6 X X m1 n1

! 0:2. reputation be

KQk m; n

Let Alberts knowledge K reputation Albert 0:8 SE K tie Steve; Albert k

KQserialized Steve; AlbertK reputation Albert k SE 0:2 0:8 0:16. As a result, the knowledge relationship tie between Steve and Albert is 0.16. Let Chriss knowledge prociency, K proficiency Chris be SE the matrix shown in Fig. 5(b), then the calculation of the degree to which Alberts prociency conforms to Steves

ARTICLE IN PRESS
S.J.H. Yang, I.Y.L. Chen / Int. J. Human-Computer Studies 66 (2008) 3650 45

request is denoted as KQSE Steve; Chris K proficiency ChrisK conformance SteveT SE SE 3T 32 2 0 0 0 0 0 0 0:2 0:7 0:7 0:7 0:8 0:8 7 76 6 6 0:8 0:7 1 0:8 0:7 0:7 76 0 0 1 0 0 0 7 7 76 6 6 7 76 6 0:2 0:2 0:3 0:2 0 0 76 0 0 0 0 0 0 7 5 54 4 0 0 0 0 0 0 0 0 0 0 0 0 3 2 0 0 0 0 0 0 7 6 60 0 1 0 0 07 7 6 6 7. 60 0 0 0 0 07 5 4 0 0 0 0 0 0 ! KQk m; n 1:0. After serialization KQserialized Steve; Chris k
4 6 X X m1 n1

change depending on how successful the interaction is. The scores of social familiarity can be updated manually as peers interact with others. Peers in the social network can explicitly update their social familiarity by changing the scores of social familiarity. Social familiarity can exhibit different levels of familiarity relationships, such as friends, teammates, organization colleagues, or virtual community members. Meanwhile, social familiarity can be positive or negative values ranging between 1 and 1, indicating bad or good relationships. In our experiment, we found that it is very difcult to obtain negative reports from testers. This phenomenon is more obvious when testers realize that the information will be used for the calculation of social relationships. Besides the fear of retaliation, people are prone to build good reputation for popularity, thus most of them would rather give a neutral response (the default value of familiarity is 0) than disclose a negative relationship. To perform quantitative analysis, we dene social familiarity between peers i and j as follows: Sfamiliarity i; j 0, if there is no relationship between peer i and peer j. Sfamiliarity i; j 0:821:0, if peer i considers peer j a friend with positive relationship. Sfamiliarity i; j 0:520:7, if peer i considers peer j a teammate with positive relationship. Sfamiliarity i; j 0:320:4, if peer i considers peer j an organization colleague with positive relationship. Sfamiliarity i; j 020:2, if peer i considers peer j a virtual community member with positive relationship. Sfamiliarity i; j 0 to 0:2, if peer i considers peer j a virtual community member with negative relationship. Sfamiliarity i; j 0:3 to 0:4, if peer i considers peer j an organization colleague with negative relationship. Sfamiliarity i; j 0:5 to 0:7, if peer i considers peer j a team-mate with negative relationship. Sfamiliarity i; j 0:8 to 1:0, if peer i considers peer j a friend with negative relationship. Each peer has a social reputation which is the product of the peers social rating (Yang et al., 2006) and the average of the peers social familiarities. Social rating indicates the degree of popularity of a target peer. It is an average rating given by other peers who know the target peer. Social reputation represents the degree of condence other peers have in a target peer. The social reputation of peer j is computed as follows: Sreputation j AVGSfamiliarity j; NoPj S rating j PM m 2 NoPjSfamiliarity j; m m1 M S rating j, where AVGS familiarity j; NoPj is an average value of peer js social familiarities, S rating j is peer js social rating, NoP(j) is a set of peers connected to j. The number of peers in the NoP(j) is M.

Let Chriss knowledge reputation be K reputation Chris SE 0:6, then K tie Steve; Chris k KQserialized Steve; ChrisK reputation Chris k SE 1:0 0:6 0:6. As a result, the knowledge relationship tie between Steve and Chris is 0.6. In summary, this example shows that Chris is more knowledgeable than Albert in terms of helping Steve apply conceptual knowledge of Software Engineering. 4.2. Calculation of social relationship tie The social relationship tie indicates the degree of social familiarity between pairs of peers on the S-net. For a pair of peers, as denoted by peer i and peer j, the social relationship tie between them is the product of their social familiarity and social reputation. S tie i; j Sfamiliarity i; j S reputation j, where Stie i; j is the social relationship tie between peer i and peer j, S familiarity i; j is the social familiarity between peer i and peer j, and S reputation j is peer js social reputation. Social familiarity indicates the level of familiarity ranging from casual to close. Each peer existing in a social network needs to specify its social familiarity with a new peer rst connecting to the social network, by lling forms and answering questionnaires. If an existing peer does not specify social familiarity with a new peer, the default value is zero meaning that there is no relationship between them. A new peer would presumably initially have a social familiarity score of 0 with all other existing peers in the social network. As it interacts with others the score will

ARTICLE IN PRESS
46 S.J.H. Yang, I.Y.L. Chen / Int. J. Human-Computer Studies 66 (2008) 3650

Please refer to Fig. 4 where you can nd all the values of the Sfamiliarity that are needed to follow the following examples. The detail calculation of social relationship tie is illustrated as follows: (1) Let Albert be a good friend of Steve, Sfamiliarity Steve; Albert 0:9. (2) Let Alberts social rating be 0.6. (3) Let Albert be directly connected to three peers (Steve, Chris, and Bob). Then Alberts social reputation is

Steve and Albert. We can also infer that Chris is more likely to share his knowledge with Steve.

5. Experiments and discussions of interactive collaboration We have conducted quantitative and qualitative experiments to evaluate the performance of our social network, P2P network, and IM group collaboration. A total of 156 undergraduate students (computer science major) participated in this experiment. They were required to install the social network-based P2P network and IM on their own computers. In order to establish the three-layer social

S reputation Albert P3 m 2 NoPAlbertS familiarity Albert; NoPAlbert Srating Albert m1 3 S familiarity Albert; Steve Sfamiliarity Albert; Chris Sfamiliarity Albert; Bob S rating Albert 3 0:9 0:8 0:2 0:6 0:3. 3 So the social relationship tie between Steve and Albert is S Steve; Albert Sfamiliarity Steve; Albert S reputation Albert, S tie Steve; Albert 0:9 0:3 0:27. (1) Let Chris be a good friend of Steve, Sfamiliarity Steve; Chris 0:8. (2) Let Chriss social rating be 0.8. (3) Let Chris be directly connected to three peers (Steve, Albert, and Bob). Then Chriss social reputation is
tie

networks, each student (peer) was asked to ll out forms and answer questions to help the system identify and calculate peers knowledge and social relationship ties. Adopting the calculation presented in this paper, a social network-based P2P network and IM is established. 5.1. Experiment To facilitate the experiment, we have developed a P2P network, called SOtella as shown in Fig. 6. SOtella is implemented based on open-source software, Edutella. In order to control the experiment scale and monitor SOtellas performance, we conne the search scope of SOtella to a

S reputation Chris P3 m 2 NoPChrisS familiarity Chris; NoPChris S rating Chris m1 3 S familiarity Chris; Steve Sfamiliarity Chris; Albert Sfamiliarity Chris; Bob S rating Chris 3 0:8 0:8 0:4 0:8 0:52. 3

So the social relationship tie between Steves and Chris is S tie Steve; Chris Sfamiliarity Steve; Chris S reputation Chris S tie Steve; Chris 0:8 0:52 0:42. This example concludes that the social relationship tie between Steve and Chris is stronger than that between

small-scale network including about 150 peers within a university. Each peer in SOtella is associated with knowledge relationship tie and social relationship tie as we presented in the three-layer social networks. We also constructed an IM-enabled group collaboration tool as shown in Fig. 7. Peers can communicate with collaborators that are able to fulll their needs and improve collaboration. From the peers found on K-net and S-net as presented in Layer-1 and Layer-2, groups can

ARTICLE IN PRESS
S.J.H. Yang, I.Y.L. Chen / Int. J. Human-Computer Studies 66 (2008) 3650 47

Fig. 6. Screen shots of Sotella.

be formed based on peers knowledge relationship tie and social relationship tie. For example, the peer Steve can request for a group formation with four members who have the strongest knowledge relationship ties to the subject Applying conceptual knowledge of Software Engineering. For adaptive group formation, peers can provide feedback to the social network via the rating mechanism and improve the calculation of the two types of tie on our three-layer social networks. 5.2. Quantitative analysis For quantitative performance evaluation, we measure two indexes: Precision and Recall. Precision is the fraction of the found peers that the requesting peers consider relevant; Recall is the fraction of the relevant peers that has been found. Precision and Recall are formally dened as follows: Precision jRaj ; jAj Recall jRaj , jRj

In this experiment, we used four kinds of knowledge domain as the search domains: Internet computing, Web computing, Mobile Internet, and Wireless Web. As indicated in Fig. 8, for the four given search domains, the Precision of social relationship tie (S-Tie) search outperforms the knowledge relationship tie (K-Tie). This indicates that the discovered peers are more relevant and they are more likely to be in the same social group because they have higher social relationship tie. In contrast, the Recall of K-Tie search outperforms S-Tie. This indicates that peers with the same knowledge domain will most likely be found at the same time because they have higher knowledge relationship tie. 5.3. Qualitative analysis Three months after the system started to run, we conducted a survey with the same 156 students who had participated in this experiment. Each student was given a questionnaire containing 14 questions to verify their satisfaction rates regarding our system. For this measure, a ve-point Likert scale with anchors ranging from strongly disagree (1) to strongly agree (5) was adopted. We then calculated the mean value and standard deviation of each question item. The result of the survey is summarized in Table 1.

where A contains a set of discovered peers, jAj is the number of peers in A, R contains a set of discovered peers who are considered relevant, jRj is the number of peers in R, Ra contains the intersection of sets R and A. jRaj is the number of peers in Ra.

ARTICLE IN PRESS
48 S.J.H. Yang, I.Y.L. Chen / Int. J. Human-Computer Studies 66 (2008) 3650

Fig. 7. IM-enabled group collaboration tool.

0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 Internet computing Web computing Mobile Internet Wireless Web

Keyword (Precision) K-Tie (Precision) S-Tie (Precision)

Keyword (Recall) K-Tie (Recall) S-Tie (Recall)

Fig. 8. Precision and Recall as performed by knowledge relationship tie (K-Tie) and social relationship tie (S-Tie).

Results of the survey revealed students doubts about the stability of the P2P search results. They could not obtain the same search results from each trial even though they used the same search option. This is due to the decentralized nature of P2P network; it only searches for the peers currently on line. Another reason of nondeterministic search results is the precaution against opportunism. Some students are not willing to share the resources they obtained from the network because free-

riders are able to acquire the same resources when they used the same search option. From the survey results, we found that most of the students were not satised with the automatic group formation. Rather, they preferred to nd their own collaborators even though they admitted that the group members picked by our system were knowledgeable. This observation suggests that in addition to knowledge competence, we should also consider users social relationships when we form groups. However, the main purpose of such a system is to generate social networks for users who do not already know other users. If the users know each other only as peers on the network and their friendship judgments are based on previous interactions rather than personal relationships then the system is more useful and this problem is not an issue. In addition, most of the students stressed the importance of user interface design of group collaboration, and they expected better controls of dynamic group formation. We also perceived that students desired powerful collaboration instruments, such as voice communication and e-whiteboard for synchronous discussion and le sharing. Our experiment conrms the effectiveness of our social network-based collaboration support to enhance knowledge sharing. Most students expressed their willingness to utilize the system for their daily studies. We also intend to use other test audiences (e.g., commercial users) to further verify the generalizability of the current ndings. In fact, many part-time graduate students in our university have showed interests in adopting our system in their daytime

ARTICLE IN PRESS
S.J.H. Yang, I.Y.L. Chen / Int. J. Human-Computer Studies 66 (2008) 3650 Table 1 Average scales of the survey of social network-based collaboration and group formation No. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 Questionnaire Are you interested in nding collaborators by using social network system? Are you satised with the system performance in terms of connection time and searching time? Can you nd the same results every time you use the P2P network? Do you keep your P2P network on line for most of the time? Are you always willing to share the resources obtained from the P2P network? Are you satised with the group formation provided by social network system? Do you think your group member chosen is knowledgeable? Do you think you can form better group by yourself? Are you satised with the user interface design of group collaboration? Is it easy to form a group? Are you satised with the group discussion performance in terms of communication and synchronization? Do you think it is important to connect to other IMs? Do you think it is important to have voice-enabled discussion? Do you think it is important to have e-whiteboard for synchronous le sharing? Mean 3.45 3.67 3.97 4.47 3.65 3.24 3.97 4.56 3.95 3.67 3.56 4.23 4.37 4.18 SD 1.34 0.80 1.29 0.75 1.75 1.13 0.78 1.10 0.94 1.28 0.84 0.62 0.68 0.61 49

Number of students 156, Strongly disagree 1, Strongly agree 5.

jobs. Some even plan to conduct small-scale experiments in their departments, which could extend the scope of the experiment to broader communities of practice with realworld cases. 6. Conclusion and future research The major contribution of this study is applying social network-based collaboration to enhance knowledge sharing in a P2P network by overcoming the difculty in nding relevant content and collaborators to interact with. So far, all the graphical representation, equations, and examples are based on direct relationship for the ease of illustration. In our future research, we will continue to expand such mathematical model of social network for more advanced reasoning of social network analysis. In the real-world society, mostly we consider individuals who have direct relationships such as Steve is a friend of Irene. Nevertheless, we also need to take into account some cases of isolated individuals and individuals with indirect (or transitive) relationships. Some individuals could be isolated and are thus disconnected from the society. On the other hand, an individual may know the friends of his/her friends. For example, if Steve is a friend of Irene, and Irene is a friend of Albert, then we might derive a transitive relationship between Steve and Albert. To fully mimic the real-world society, a social network comprising both direct and transitive relationships should be modeled as a non-connected graph (network). So far, all graphical representation, equations, and examples presented in this paper are based on connected graph and direct relationships. In our future research, we will continue to expand the mathematical model from connected graph to non-connected graph, and continue to develop new computing mechanism for deriving transitive relationship. We plan to continue our research work in the following two directions. First, it is a general problem for a social

network to support the discovery, access, and sharing of knowledge. Users and other collaborators may have their own needs when they access subjects and discuss with others. We plan to conduct further study to investigate the special requirements from different social networks or social relationships in virtual communities. Second, to take into account the context of collaboration, we plan to explore applied social networks to combine collaborative domain and collaborators ontology.

Acknowledgments We would like to express special thanks to the anonymous reviewers for their thoughtful comments and constructive suggestions. Many of the ideas and wording shown in this paper come from the reviewers comments. This work is supported by the National Science Council, Taiwan under Grants NSC 95-2520-S-008-006-MY3 and NSC 95-2416-H-231-010.

References
Aberer, K., Punceva, M., Hauswirth, M., Schmidt, R., 2002. Improving data access in P2P systems. IEEE Internet Computing 6 (1), 5867. ACM Computing Classication System, 1998. Available from /http:// www.acm.org/class/1998/S. Alexander, P., 2005. Should Your Business Use Instant Messaging? /www.msnbc.comS. Anderson, L.W., Krathwohl, D.R., Airasian, P.W., Cruikshank, K.A., Mayer, R.E., Pintrich, P.R., Raths, J., Wittrock, M.C., 2001. A Taxonomy for Learning, Teaching, and Assessing: A Revision of Blooms Taxonomy of Educational Objectives. Longman, New York. Barak, M., Rafaeli, S., 2004. On-line question-posing and peer-assessment as means for web-based knowledge sharing in learning. International Journal of HumanComputer Studies 61 (1), 84103. Bieber, M., Engelbart, D., Furuta, R., Heltz, S.R., Noll, J., Preece, J., Strhr, E.A., Turoff, M., van de Valle, B., 2002. Toward virtual community knowledge evolution. Journal of Management Information Systems 18 (4), 1135.

ARTICLE IN PRESS
50 S.J.H. Yang, I.Y.L. Chen / Int. J. Human-Computer Studies 66 (2008) 3650 Hill, R., Dunbar, R., 2002. Social network size in humans. Human Nature 14 (1), 5372. Huhns, M.N., Buell, D.A., 2002. Trusted autonomy. IEEE Internet Computing, 9295. Isaacs, E., Walendoski, A., Whittaker, S., Schiano, J.D., Kamm, C., 2002a. The character, functions, and styles of instant messaging in the workplace. In: Proceedings of ACM 2002 Conference on ComputerSupported Collaborative Work (CSCW02), New Orleans, LA. Isaacs, E., Walendowski, A., Ranganathan, D., 2002b. Hubbub: a soundenhanced mobile instant messenger that supports awareness and opportunistic interactions. In: Proceedings of ACM Conference on Human Factors in Computing Systems CHI 02, Minneapolis, MN, pp. 179186. Jin, Q., 2002. Design of a virtual community based interactive learning environment. Information Sciences 140 (12), 171191. Kautz, H., Selman, B., Shah, M., 1997. Referral Web: combining social networks and collaborative ltering. Communications of ACM 40 (3), 2736. Mitchell, T., 1997. Machine Learning. WCB McGraw-Hill, New York, pp. 128141. Nardi, B., Whittaker, S., Bradner, E., 2000. Interaction and outeraction: instant messaging in action. In: Proceedings of ACM 2000 Conference on Computer-Supported Collaborative Work (CSCW00), Philadelphia, PA. Nejdl, W., Wolf, B., Qu, C., Decker, S., Sintek, M., 2002. Edutella: a P2P networking infrastructure based on RDF. In: Proceedings of the 11th International World Wide Web Conference, May 2002, Hawaii, USA. Polanyi, M., 1997. The tacit dimension. In: Prusak, L. (Ed.), Knowledge in Organizations. Butterworth-Heinemann, Woburn, MA. Raghavan, P., 2002. Social networks: from the Web to the enterprise. IEEE Internet Computing 6 (1), 9194. Ravishankar, C.V., 2002. The Gnutella Protocol Specication Version 0.4. Available from /http://www.clip2.comS. RDF, Resource Description Framework. Available from /http:// www.w3.org/RDF/S. Wachter, R.M., Gupta, J.N.D., Quaddus, M.A., 2000. It takes a village: virtual communities in supporting of education. International Journal of Information Management 20 (6), 473489. Wasko, M.M., Faraj, S., 2005. Why should I share? Examining social capital and knowledge contribution in electronic networks of practice. MIS Quarterly 29 (1), 3557. Yang, S.J.H., Hsieh, J.S.F., Lan, B.C.W., Chung, J.Y., 2006. Composition and evaluation of trustworthy Web services. International Journal of Web and Grid Services 2 (1), 524. Zhang, Y., Tanniru, M., 2005. An agent-based approach to study virtual learning communities. In: Proceedings of the 38th Hawaii International Conference on System Sciences. Bistrom, J., 2005. Peer-to-peer networks as collaborative learning environments. In: Paper Presented at HUT T-110.551 Seminar on Internetworking. Bloom, B.S., Engelhart, M.B., Furst, E.J., Hill, W.H., Krathwohl, D.R., 1956. Taxonomy of Educational Objectives: The Classication of Educational Goals, Handbook 1: The Cognitive Domain. Longmans Green, New York. Brase, J., Painter, M., 2004. Inferring metadata for a semantic Web peerto-peer environment. Educational Technology & Society 7 (2), 6167. Bruckman, A., 2002. The future of e-learning communities. Communications of the ACM 45 (4), 6063. Carroll, J.M., Neale, D.C., Isenhour, P.L., Rosson, M.B., McCrickard, D.S., 2003. Notication and awareness: synchronizing task-oriented collaborative activity. International Journal of HumanComputer Studies 58 (5), 605632. Chau, R., Yeh, C.H., 2004. Fuzzy Conceptual indexing for concept-based cross-lingual text retrieval. IEEE Internet Computing 8 (5), 1421. Chen, I.Y.L., Yang, S.J.H., 2006. Peer-to-peer knowledge sharing in collaboration supported virtual learning communities. In: Proceedings of the Sixth IEEE International Conference on Advanced Learning Technologies (ICALT), Netherlands, July 2006, pp. 807809. Churchill, E.F., Halverson, C.A., 2005. Social networks and social networking. IEEE Intelligent Systems 20 (5), 1419. Diamadis, E.T., Polyzos, G.C., 2004. Efcient cooperative searching on the Web: system design and evaluation. International Journal of HumanComputer Studies 61 (5), 699724. Edutella. Available from /http://edutella.jxta.org/S. Eikemeier, C., Lechner, U., 2003. Introducing domain specic ad-hoc collaboration: the peer-to-peer tool iKnow. In: Proceedings of the 12th IEEE International Workshops on Enabling Technologies: Infrastructure for Collaborative Enterprises (WETICE03). Fischer, G., Giaccardi, E., Eden, H., Sugimoto, M., Ye, Y., 2005. Beyond binary choices: integrating individual and social creativity. International Journal of HumanComputer Studies 63 (45), 482512. Gall, M.D., 1987. Discussion method. In: Dunkin, M.J. (Ed.), The International Encyclopedia of Teaching and Teacher Education. Pergamon, Oxford. Gnutella. Available from /http://www.gnutella.com/S. Grinter, R., Palen, L., 2002. Instant messaging in teen life. In: Proceedings of ACM 2002 Conference on Computer-Supported Collaborative Work (CSCW02), New Orleans, LA. Groove, 2006. Available from /http://www.groove.net/S. Herbsleb, J.D., Atkins, D.L., Boyer, D.G., Handel, M., Finholt, T.A., 2002. Introducing instant messaging and chat in the workplace. In: Proceedings of ACM Conference on Human Factors in Computing Systems CHI 02, Minneapolis, MN.